as far as I remember, Martin presented a few years ago an improvement **in terms of success rate** for a method with a slight bonus **for >0.5 margin wins** in Fuego. I don't remember the details... Olivier
2013/5/31 Woody Folsom <[email protected]> > This happens because most MCTS engines assign the same value (reward) to a > terminal board state regardless of the margin: e.g. 1 point for a win and 0 > points for a loss. Thus, my program would pick a move that has an > estimated 75% chance of winning by an average of 0.5 points over one that > has an estimated 74% chance of winning by 1.5 points. One could change > this behavior by assigning a higher reward to a winning board state with a > larger margin, but then a competent player could draw the computer into > making very risky moves that appear to potentially have a huge payoff. > > Woody > >
_______________________________________________ Computer-go mailing list [email protected] http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
